AI Tech Daily May 13, 2026

1. Executive Summary

On May 13, 2026, the AI industry is focusing on advancements in simulation technology to accelerate scientific discovery and the evolution of human-AI interaction interfaces. Microsoft’s research group is redefining AI’s role in materials science, while Google DeepMind is rebuilding UI, unchanged for half a century, from an AI perspective. Furthermore, in the corporate world, NVIDIA and SAP are collaborating to establish a framework for ensuring the safety and governance of autonomous AI agents, marking a new stage in AI implementation.

2. Today’s Highlights

Microsoft’s Materials Science Platform “MatterSim” Validates Predictions Through Experiment

Microsoft Research has announced the latest update to its AI-powered materials discovery platform, “MatterSim.” The highlight of this update is the successful experimental synthesis of a new material called “orthorhombic tantalum phosphide” (TaP), which MatterSim had predicted. This signifies the physical realization of computational scientific predictions in a real laboratory, strongly indicating the potential to significantly shorten the materials development cycle. Compared to traditional first-principles calculation methods, MatterSim can perform inference several times faster, and this latest update further improves simulation speed by 3-5 times. This enables the modeling of complex multi-property phenomena, with applications expected across a wide range of fields, from energy storage to nanoelectronics. The technological significance lies in AI taking on a decisive role as a “discovery partner” in new materials development, which has traditionally required immense time and cost. In terms of industry impact, the pace of development for next-generation thermal conductors and energy-efficient devices could accelerate, potentially building an overwhelming advantage in hardware competition. Source: Microsoft Research Official Blog “Advancing AI for materials with MatterSim”

Google DeepMind Redefines the “Pointer” for the AI Era

Google DeepMind has announced research into redesigning the traditional mouse cursor (pointer) to make AI and human interaction more seamless. While the pointer has maintained its role for over half a century in PC history, DeepMind researchers are exploring a pointer that understands the user’s workflow and infers not only “what” is being pointed to on the screen but also “why.” For example, an AI that understands the content of an image on the screen and intuitively interacts with a map tool simply by the user issuing a command like “show me the route.” This is a crucial step towards eliminating the fragmentation of “AI existing in another window” and directly integrating AI into any application the user employs. The evolution of interface technology will undoubtedly accelerate the adoption of AI, and in the future, how to intuitively perform “collaborative work with AI,” not just “conversations with AI,” will determine the success or failure of product design. This research is expected to be deployed to other platforms as foundational technology for future generative UI. Source: Google DeepMind “Reimagining the mouse pointer for the AI era”

3. Other News

• NVIDIA and SAP Enhance Governance for Enterprise AI Agents NVIDIA and SAP have announced a collaboration to enable enterprises to deploy autonomous AI agents safely and reliably. SAP will integrate NVIDIA’s open-source runtime, “OpenShell,” into its business AI platform. This will allow for the enforcement of security policies when AI agents operate in enterprise environments, enabling advanced automation with assured governance. Source: NVIDIA Official Blog

• Microsoft Releases Benchmark to Measure “Social Behavior” of AI Agents Microsoft has released “SocialReasoning-Bench,” a benchmark for evaluating the social reasoning capabilities of AI agents. This metric measures “how well agents can represent user interests” in tasks requiring collaboration with others, such as calendar scheduling and market negotiation. It has been revealed that while many frontier models can perform the tasks themselves, they tend to fail to maximize user interests during negotiations. Source: Microsoft Research Official Blog

• Meta Expands Age Restriction Enforcement Using AI Meta is expanding the deployment of AI technology that detects age restriction violations from the context of posts and comments on Facebook and Instagram. Unlike facial recognition that identifies users, this technology estimates age using general visual and contextual cues such as posture and height, with the aim of enhancing platform safety. Source: Meta Official Newsroom

• OpenAI Rolls Out “GPT-5.5 Instant” to All Users OpenAI has updated the default model for ChatGPT to “GPT-5.5 Instant.” This model offers up to a 52.5% reduction in hallucinations (misinformation) and significantly improved reasoning capabilities in STEM domains compared to its predecessor, 5.3. It will be gradually rolled out to all ChatGPT users. Source: OpenAI Blog

• Anthropic Announces “Dreaming” Feature for AI Agents Anthropic has announced a “dreaming” feature that allows AI agents to review their past sessions between jobs and generate insights for improvement. Pilot tests with partner companies like Harvey have shown results, with task completion rates improving by approximately six times. Source: Anthropic Official Blog

4. Summary and Outlook

Looking at the overall news today, it’s clear that AI technology is rapidly shifting from mere generative capabilities to “reasoning,” “experimentation,” and “autonomous workflow execution by agents.” In particular, the fact that AI can now achieve results with physical world implications, as demonstrated by Microsoft’s MatterSim, represents a major paradigm shift for industries. On the other hand, as seen in the partnership between NVIDIA and SAP, governance frameworks for enterprises to safely operate such advanced agents are an urgent issue. The coming months will see intensified competition not only in the performance improvements of new models but also in the infrastructure layer, focusing on “how to safely and controllably implement autonomous agents into society.”

5. References

This article was automatically generated by LLM. It may contain errors.